Chapter 13Ninety Days: Complete the First Cycle of Organizational Evolution

1. First knock down the “three-year project” #

Speak of organizational transformation and most principal leaders reach for two moves: hire a consulting firm for top-level design, and stand up a three-year plan. I oppose both. The reason is 10/10/80: extracting standards scores 10, collecting data scores 10, seeking the better solution by means scores 80. Top-level design, however perfect, covers only the first 20. A three-year plan’s real function is often to postpone the 80 of execution until “we are ready to start.” No standard, stop execution. But the reverse is also true: standards written and still not executed, and planning becomes the most respectable form of delay.

The problem with consultants is the structure of the engagement, not their professional skill. Transformation’s core deliverable is the eight-layer standard, and the first author of those standards can only be the principal leader. Writing standards is itself the most expensive learning in the transformation: every “condition + action + acceptance” forces a once-tacit judgment into the open. Outsource that process and you buy a handsome document while missing the only move that actually rewrites the gene. Outsourcing the most expensive learning is paying someone else to grow your muscle. Advisors may run alongside, interrogate, and hand tools; they may not ghostwrite.

The alternative I offer is a ninety-day minimum closed loop: set standards → build consensus → collapse one execution unit. What the loop can finish is the first cycle of organizational-DNA evolution: the organization’s highest-frequency judgments have standards; core members think in one language; at least one execution unit has finished an AI rebuild and come back alive. A finished transformation in ninety days would be a lie. Organizational DNA is the default of judgment, so ninety days cannot change everything about the company, but it is enough to change what the company believes.

Three months is the early-company verification window in the people-investment standard. Organizational transformation is itself an investment, and it must accept the same discipline as every other investment: freeze acceptance criteria before spending, settle with bilateral evidence at term. Managing transformation as a ninety-day investment is itself the new gene’s first expression.

2. The ninety-day roadmap #

Four segments, thirteen weeks. Each segment gets three things: moves, deliverables, most common deaths.

Fit scale: this roadmap is designed for organizations of twenty or fewer core members, with the premise that the principal leader can personally run Feynman acceptance for each. Larger organizations need not change the roadmap; change the scope: treat one division or independent business unit as “the company,” and let that unit’s principal leader walk all thirteen weeks. Once the unit runs, it becomes the group’s internal Transn—a rebuild prototype you can visit and calibrate against, more persuasive than any top-down all-hands. A retestable sample sits at the top of the credibility ladder.

Weeks 1–2: Clarity Method audit #

The move is Clarity Method at organization scale: change nothing yet; first widen your own options. Three lists:

1. Limiting-resource inventory: which resource is the company stuck on—cash, judgment, or consensus? Verify by where money flows, not by meeting minutes: declarations manage other people’s judgments; cash flow exposes your own.
2. High-frequency judgment count: which judgments were repeated in the organization over the past month? Rank by frequency; take the top three—they are the objects of standardization in weeks 3–6.
3. User-consensus location: which tier of the adoption curve do your paying users sit on? Is your product path forward or reverse?

Deliverable: a one-page status map. Most common death: the audit becomes a mobilization rally. Announcing any change in weeks 1–2 is jumping the gun—you have no standards yet; the louder the announcement, the higher the later cost of walking it back.

Weeks 3–6: standards engine starts #

Three core moves:

1. Write the eight-layer standard (mother-table structure): mission, belief, behavior, thinking, judgment, people investment, management, learning—in that order. Do not seek perfection; seek executability. Each clause in three parts—“condition + action + acceptance”—better rough and decidable than elegant and ambiguous.
2. Build ROI cards: personal ROI cards for every member; verification windows of 3, 6, or 9–12 months by organizational stage; before investment freeze value definition, full cost, attribution method, and stop conditions.
3. Standardize the top three judgments: write the three highest-frequency judgments from weeks 1–2 into executable standards. One technical detail: use the median for statistics. Judgment samples always hold extremes; the mean is hijacked by them; the median reflects how the typical case should be judged.

Deliverable: eight-layer standards v1 plus org-wide ROI cards. Two symmetric common deaths: perfectionism (still rewriting layer-one wording in week 6) and the democracy trap (draft standards put to an all-hands vote). First author of the eight-layer standard can only be the principal leader; consensus is manufactured, not voted.

Weeks 7–10: consensus engine runs #

Standards on paper are not yet gene; gene is when they enter everyone’s default judgment. Three moves:

1. Internal Interference Method: give critical standards short names so they become daily language—only standards that can be quoted offhand get executed offhand. The principal leader says them ten times; every layer says them ten times more; and place standards where anyone can reach them at a touch. Shared context comes from infrastructure that makes fetching easier than hoarding, not from exhortation.
2. Feynman acceptance (learning standard): one by one, have each core member explain the eight-layer standard to a newcomer in simple language. Where explanation fails, first assume the standard is badly written, then assume the person has not understood—and fix both on the spot.
3. Set the product path: on the user-consensus location from weeks 1–2, formally choose the forward path or consensus reverse iteration; write the choice into the R&D roadmap with the five gates.

Deliverable: every core member passes Feynman acceptance; one-page product-path resolution. Most common death: treating “we sent an all-hands email” as “we built consensus.” Consensus acceptance is that others can judge in your language. Receipt of the language is not enough.

Weeks 11–13: dual pilots #

The first ten weeks lay track; the last three move for real. Two pilots open together:

1. Execution-unit collapse: pick the execution unit densest in S1/S2; finish AI or outsourcing rebuild; re-grade remaining roles; reconfigure each person’s rung on the decision-rights ladder. Selection criteria: small enough (failure survivable), real enough (live business traffic), painful enough (savings visible to finance).
2. Single-variable product iteration: pick one product unit; run one single-variable iteration under progressive R&D standards—name the one uncertainty to eliminate now, keep a stable core and rollback, update the roadmap from results.

Each pilot gets a stop line written in advance—the “no standard, stop execution” iron rule in pilot form. In the execution-unit rebuild: if delivery quality falls below the pre-rebuild waterline and cannot be attributed within two weeks, pause and roll back. ’s experience: a first-month dip is unsurprising, but the dip must attribute to a concrete standards gap; unattributable chaos may not be endured. In product iteration: any damage to a stable core’s existing metrics triggers rollback—no exceptions. Stop lines bind only if written and published before launch. Lines drawn mid-pilot, wherever drawn, look like steps built for failure.

Deliverable: one pilot review with bilateral evidence complete. Succeeded—evidence enters the standards library as clauses. Failed—attribution clear (bad standard, drifted execution, or thin context), also into the library. R&D progress is counted in how much critical uncertainty was eliminated, not in new features; transformation progress the same.

3. Eight-question acceptance #

At ninety days’ end, do not accept on “how many AI tools we deployed”—a tool list is the easiest fake report card. Accept on eight questions:

1. Can core members explain the eight-layer standard in simple language?
2. Can the last three important decisions be traced to an explicit value judgment?
3. Does every member hold an ROI card matched to organizational stage, and do leading indicators support ROI greater than one at term?
4. At which decision-rights rung does each core member sit, and are upgrade conditions clear?
5. Does daily reflection keep producing new standards, revisions, or better solutions?
6. Do AI and humans act in the same context under the same standards?
7. Does R&D name current user consensus, product path, and the single uncertainty under test—with stable core and rollback retained?
8. When experience conflicts with a standard, do members pause on their own, seek bilateral evidence, and revise judgment—rather than appeal to seniority or habit?

The eight questions map the book’s skeleton: one tests the standards engine, two value judgment, three investment discipline, four decision-rights migration, five the learning loop, six shared context, seven product path; eight tests the gene itself. Organizational DNA is the judgments and actions that still happen when no one is watching, not values on the wall. Pass six of eight and the first cycle is complete; under four, return to weeks 1–2 and re-audit—most likely the limiting-resource call was wrong.

One discipline of method: the principal leader asks others the eight questions, never themselves. Self-tested eight questions always pass. Feedback avoidance in acceptance form is handing paper and answer key to the same person. Operationally, draw three core members at random through the eight; take the worst score. Where three answers contradict on one question, count fail—because the eight test consensus, and consensus fails by divergence rather than by a single error. The most common crash in spot checks is question eight, which no questionnaire can ask; you must wait on events. The next scene where experience conflicts with a standard, watch whether the first reflex is to open the standard or to invoke seniority. That instant’s answer is more honest than every report in the ninety days combined.

4. A fully assembled machine: the bioby.ai case #

Load every standard the book has named back into one company and you see how they mesh. This is how my own company runs every day, not a showcase case. The book’s eight-layer mother table was lifted from here. The machine’s outward business is an overseas influencer-marketing Agent, chosen by the variable × invariant multiplication; the machine and the business it processes run on the same set of standards.

Assembly order is logical order. At the top, mission: use AI to raise the level of all humanity; the company name bioby is bio by ai; the logo is the letter b built from the digits 1 and 0 (a computer’s substrate is 0 and 1; this machine’s substrate is Clarity Method and Interference Method; the logo locked in two days). Beneath mission, belief: human–machine collaboration is humanity’s future. Neither layer directs any concrete judgment, but they answer the hardest of the eight: unsupervised, why trust that members’ judgments still point the same way. Next, one behavior standard—treat people with sincerity—the ethical boundary of Interference Method and the trust foundation of the consensus engine; one judgment standard—judge by value—the company version of value judgment. Below that, four layers of operating system: people-investment standard (org-wide ROI greater than one, staged verification), management standard (five-rung ladder from instruction to delegation), learning standard (reflection plus Feynman), role structure (S1/S2 outsourced; S3 and above all think) plus time allocation (1–2 hours versus 6).

Facing the ninety-day roadmap, the question readers should ask me most is: how many days did you yourselves walk? The honest answer is longer than ninety, because we had no manual; every step hit a wall first, then derived. Timeline open (internal company record): February 2026 derived Clarity Method—the first cornerstone. Before that, hiring, fundraising, and logo design all ran as execution without standards; tuition booked to three ledgers—the business-role funnel, fundraising rework, and logo idle spinning. April derived Interference Method; the same month the four hiring standards landed; first move was to recalibrate every sitting role—a standard’s first users are always the stock. March to July, the R&D unit finished collapse rebuild; ledger already open: cost about 2.3×, capacity about 5×, bug rate down about 80%. May engaged a professional FA (stop-loss on the fundraising tuition); time-allocation recording started the same month. June, ROI > 1 was forced out of a people-budget crunch; logo standard locked in two days; meeting-room renovation paused under the iron rule. From first cornerstone to the eight-layer mother table roughly meshed took about five months. Those five months were two overlapping cycles: set standards and build consensus laid across February–April; collapse pilot and investment discipline started in March and closed in July. The ninety-day manual is the straight line left after cutting every detour from those five months. With the map in hand, you should walk faster than the people who drew it.

That timeline hides a structure worth naming: five standards, five births, none from a plan. Clarity Method from the pain of repeated rework; the four hiring lines from tuition on twenty business-role hires; ROI > 1 from a budget that would not stretch; the logo standard from four designers idling two months; the stop-execution iron rule from muscle memory on a ten-thousand-yuan meeting room. Organizational standards are not promulgated from heaven; they are forced out when the method hits a jam. That is my last rebuttal to “do three years of top-level design first”: a standard’s birth certificate names a concrete pain. No service can feel a company’s pain for it, and no advisor can complete the principal leader’s judgment. What service can do is shorten the distance from pain to standard—help you name the limiting resource, freeze acceptance criteria, organize the first verification round, and compress five months of detours into ninety days. That is also the boundary of Regenic.ai enterprise service: accompany the rewrite; do not outsource the transformation for you.

The R&D rebuild’s structure and ledger are already open in the collapse sample—company operating figures, unaudited. The time-allocation execution record accumulates from May 2026 across seven core-team members, self-report plus one-level review; add after the first full quarter 【pending: quarterly data】. I do not substitute adjectives for numbers. Until data is in, this record is mechanism description only.

bioby.ai is early-stage; these standards run and revise daily on a small sample, but they have not faced thousand-person stress. Their role in this book is an assembly diagram. Proof goes to time—and to the two external samples below.

5. Two paths, one endpoint #

External evidence gives two symmetric paths.

Native path: Dance with Love (Yuaiweiwu) (assembly view). Founded May 2023 already designed for human–machine collaboration; product, R&D, design, marketing ops, and sales—all five core roles rebuilt for collaboration; cross-department shared data pools forming a data flywheel. In two years: four rounds of financing about $150 million, valuation near $1 billion, monthly revenue in the tens of millions of yuan (founder disclosure to 36Kr; unaudited). It never “transformed,” because it was born this species.

Rebuild path: Transn (assembly view). A translation company nearly twenty years old, sitting at the center of an industry where execution price approaches zero, remaking itself by three mechanisms: a CAIO and AI Native decision committee (reweighting judgment), a hard DEMO rule (making standards executable), and an energy-gold mechanism (aligning incentives to the new gene). This book cites mechanism only; effect data lacks third-party verification.

A newborn company and an old company, opposite starts, same destination: unified judgment standards plus unified context. That is the confidence behind the ninety-day plan: an AI-native organization is an old species that dares to rewrite its own standards. Native companies save only the cost of tearing down the old; what must be built is identical.

Beyond the two living roads sits a road already walked badly; its signpost stands here too. Klarna and MD Anderson, reread in today’s language, are two inverses of the ninety-day loop. Klarna treated transformation as a cost-cutting campaign, with cost as the only evaluation—judgment standard in the eight layers written wrong, then executed at full speed. MD Anderson treated transformation as a procurement project, bought sixty million dollars of top tools, governance bypassed standards, machines could not read context—equal to skipping weeks 3–10 of the roadmap and jumping from purchase to acceptance. Neither death has a word to do with technology. Transformation failure is almost always standard death. So in the roadmap, tools appear after standards and consensus: the order is what the autopsy reports require.

6. Boundaries of the claim #

The claim has three boundaries.

First, ninety days completes the first cycle, not the whole. Gene evolution is a loop: first-cycle deliverables already contain the next cycle’s start (new limiting resources exposed by the pilot review). Reading this chapter as “transformation done in ninety days” treats the first iteration as the final release. Most organizations need three to five cycles before the new gene expresses stably—that is one to one-and-a-half years, still far faster than “three years of top-level design,” because every cycle settles in live business.

Second, ninety days has one irreplaceable premise: the principal leader enters the field personally. First author of the eight-layer standard must be the principal leader; first Feynman examinee should be too. By the 1–2/6 accounting, that means at least two to three hours a day across the ninety on standards and consensus. If your calendar cannot yield that space, do not start: outsourcing transformation to a deputy or advisor is handing gene surgery to someone else—the output will be rejection.

Third, sample quality. Dance with Love’s financing, valuation, and revenue are founder media disclosure and company figures, unaudited; Transn cites mechanism only, not effects; bioby.ai is assembly diagram, not proof of success—internal quantitative record 【pending】. Still missing is the most persuasive sample: a mid-size traditional enterprise that completed the ninety-day loop with before/after operating data. After this book publishes, I hope such a sample appears among the first readers, and—with the company’s permission—that process, boundaries, and before/after data are recorded on Regenic.ai. Showing only success turns the method back into divination; stop lines and failure samples must remain. That is not courtesy: what I am running on you in this book is Interference Method, and its acceptance standard is the same as question eight of the eight—after you close the book, unsupervised, will you act by these standards?

What to Do Monday Morning (principal-leader view) #

Tear section two off and put it on the wall. Monday morning, do the first act of weeks 1–2: pull the last three months of spend flow and answer one question—does where the money actually goes match the limiting resource you say with your mouth?

Audit begins in honesty; transformation begins in audit.

Closing: the last scarcity #

At the end of the book, settle the account of the name “the last scarcity.”

The price of execution is approaching zero—that is the book’s starting point. Judgment becoming the last scarcity—that is the book’s body. After the standards engine turns judgment too into an asset that can be copied, iterated, and inherited, what remains scarce?

What remains is choosing what to believe. Standards can be copied; the belief that wrote the first standard cannot. Context can be unified; the sense of direction for where to unify cannot. AI took execution and is learning to assist judgment, but it has no wants—and every organizational evolution begins with someone wanting something that does not yet exist. That is the scarcity machines can never replace, and the true meaning of “human–machine collaboration is humanity’s future”: machines responsible for everything that can be standardized, and humans responsible for the next standard.

The last scarcity was never on the machine’s side. It is in the moment you decide what to do Monday morning.

Chapter Acceptance Self-Check (against the chapter’s five acceptance standards) #

1. Claim restatable in one sentence ✓, and assembly closure of the book’s argument chain (each move points back to prior standards).
2. Whiteboard framework figure ✓ (ninety-day roadmap four segments thirteen weeks; playbook cover image).
3. External comparison and data ✓: Dance with Love (native path, company caliber marked) + Transn (rebuild path, mechanism only) + Klarna/MD Anderson (transformation-failure control, Chapter 2 echo) + bioby.ai assembly diagram and five-month assembly timeline (v1.2 incorporated, quality boundary stated, settled in Chapter 10, quarterly data 【pending】 marked honestly); mid-size traditional enterprise before/after data gap marked and converted to open invitation.
4. Fifteen quotable-line candidates ✓ (v1.2 map, birth certificate, outsourced learning, standard death, test-paper five).
5. “What to Do Monday Morning” principal-leader view ✓ (this chapter is the checklist + first action concretized).
6. Fluency ✓: whole-sentence rewriting and English breath under current prose-standard.